Stereoscopic Image Retargeting And Its Quality Assessment

With the advancement of stereoscopic image and video technology and the continu-ous improvement of international 3D imaging standards,human beings are opening up a new era of 3D vision applications.This development of image processing technology and the diversification of display devices makes viewing stereoscopic content on a va-riety of devices an urgent need.In order to adapt to different viewing needs,the size of the image needs to be adjusted to achieve a good display effect on different termi-nal devices.The simplest method of image size adjustment is uniform scaling,but this method tends to cause distortion of important objects in the image,which in turn affects viewing comfort.Compared with traditional 2D visual content,stereoscopic visual con-tent introduce an additional dimension of depth,which brings more enjoyment to the users.However,this additional dimension would also bring some additional challenges and constraints for the quality of experience of stereoscopic visual content.Therefore,it is necessary to achieve image size adjustment under the premise of avoiding visual distortion as much as possible,preserving the depth and obtaining an optimal image·quality perception;this process is called stereoscopic image retargeting.However,since the current stereoscopic image retargeting technology is not mature enough,the stereoscopic image retargeting operators introduce different distortions to the image quality,depth perception and visual comfort.Distorted stereoscopic images can seriously affect user's visual of experience and even cause some visual health problems.Therefore,the quality evaluation of retargeted stereoscopic images plays a crucial role in the application of retargeted stereoscopic images and the quality of experience of the users.Exploring how to establish an objective stereoscopic image retargeting quality evaluation model consistent with the subjective perception of the human eye is a research hotspot in the field of stereo image processing.The research on stereoscopic image retargeting quality evaluation methods has important theoretical and practical significance for promoting the application and popularization of stereo image technology.Far less effort has been put forth in evaluating the retargeted stereoscopic images.A few methods have leveraged 2D image retargeting quality assessment methods to address this issue.However,these methods cannot effectively detect image structure distortion,which is a crucial factor for quality degradation of retargeted stereoscopic images.Moreover,these studies have not consider the comprehensive perspective of stereoscopic image retargeting quality.Furthermore,another severe issue in this research area is the lack of stereoscopic image quality assessment database.In this thesis,the stereoscopic image retargeting techniques and its influence on image quality are analyzed and organized.Furthermore,the key issues of quality evaluation of stereoscopic image retargeting are deeply studied,and some new ideas and methods are proposed.The main research work and contributions are as follows:First,we evaluate the stereoscopic image retargeting techniques subjectively.We start by building a stereoscopic image retargeting database with human subjective scores,then we analyze the obtained database by taking into consideration the retarget-ing methods,image content and the different quality aspects.This will provide insights and ground truth to the objective quality assessment methods.Next,we propose a novel objective quality assessment method for stereoscopic im-age retargeting by exploiting a triangulation-based method to quantitate the distortion and image quality.In term of image quality,we extract SIFT feature points from both source and retargeted images and match the corresponding point pairs.Then,we calcu-late a saliency map to filter unconfident points out.The left points are connected to form a triangular mesh which is used to quantitate the image quality by measure the distortion of each triangle pair.In addition,we quantitate the visual comfort and depth perception to constitute the feature vector which is token as input of the Support Vector Regression(SVR)model.The proposed assessment model is effective and comprehensive,and the experimental results show that the stereoscopic image quality scores predicted by this model is highly consistent with human mean opinion scores(MOS).Then,we propose a novel stereoscopic image retargeting method based on seam carving.First,a saliency map is computed for the source image.The computed saliency map is used to compute the distribution of the salient objects in the image.Next,the image is segmented into small rectangles according to the distribution of the salient ob-jects.Then the traditional seam carving is applied to each small rectangle independently.The experimental results showed that the performance of our method outperforms the traditional seam carving and other stereoscopic image retargeting methods.In conclusion,this thesis takes in-deep researches on the issues of stereoscopic im-age retargeting methods and the quality assessment metrics of stereoscopic image re-targeting,proposes the corresponding technical solutions and methods for stereoscopic image retargeting and its quality assessment A large number of experiments are con-ducted on the proposed stereoscopic image retargeting database and other 2D image retargeting databases.The results show that the proposed methods achieve a good per-formance,and outperforms the current existing methods.